[Graph] Re-add SM110 support, and use Toolkit 12.8 for SM90,100,120#780

Merged

hughperkins merged 16 commits into

mainfrom

hp/readd-sm110-multitoolkit

Jul 15, 2026

Commits on Jul 13, 2026

[CUDA] Re-add sm_110 (Thor) via per-toolkit fatbins
hughperkins
committed
[Docs] List supported CUDA architectures in supported_systems
hughperkins
committed
[Docs] Express supported CUDA archs as a bounded range
hughperkins
committed
[Test] Add fatbin arch-partition regression test
hughperkins
committed
[Build] Use an explicit per-SM toolkit mapping for fatbins
hughperkins
committed
[Misc] Apply pre-commit formatting (black + clang-format)
hughperkins
committed
[Build] Cover pre-Hopper sm_60/sm_70 in the checkpoint yield-check fatbin
hughperkins
committed
[Test] Verify embedded build toolkit per cubin, not just arch partition
hughperkins
committed
[Build] Drop the unused per-fatbin sizes table from generated headers
hughperkins
committed
Merge branch 'main' into hp/readd-sm110-multitoolkit
hughperkins
authored
[Docs] Unwrap fatbin build-guide prose and mark it Advanced:
hughperkins
committed
[Docs] Clarify why any 13.x toolkit satisfies the sm_110 >=13.0 pin
hughperkins
committed
[Build] Bound >=X.Y toolkit resolution to the same CUDA major
hughperkins
committed
[Doc] Link qd.simt.subgroup to subgroup.md in supported_systems
hughperkins
committed
[Doc] State AMD subgroup width as user-facing fact, not codegen mechanics
hughperkins
committed
[Doc] Define/avoid specialized terms in supported_systems notes
hughperkins
committed